-
Notifications
You must be signed in to change notification settings - Fork 4.3k
All issues
Issue creation is restricted in this repository
Issues
is:issue state:open
is:issue state:open
Search results
[GTP][Feat] GTP schedule optimization for MoE layer: one-block-ahead prefetch
enhancementNew feature or requestNew feature or requestStatus: Open.#6046 In NVIDIA/Megatron-LM;- Status: Open.#6037 In NVIDIA/Megatron-LM;
🐛 Weekly regression: Mixtral 8x22B loss divergence and iteration-time slowdown
bugSomething isn't workingSomething isn't workingStatus: Open.#6034 In NVIDIA/Megatron-LM;Reproducible local gradient NaN after approximately 100 steps when distilling a pruned Qwen3-8B-Base with NeMo 25.09
bugSomething isn't workingSomething isn't workingwaiting-on-customerWaiting on the original author to respondWaiting on the original author to respondStatus: Open.#6018 In NVIDIA/Megatron-LM;Add checkpoint interoperability tests between MLM and Megatron-Bridge formats
enhancementNew feature or requestNew feature or requestStatus: Open.Decide and implement tokenizer asset compatibility
enhancementNew feature or requestNew feature or requestStatus: Open.Save and load train_state.pt in MegatronLM checkpoints
enhancementNew feature or requestNew feature or requestStatus: Open.Load run_config.yaml into MegatronLM args
enhancementNew feature or requestNew feature or requestStatus: Open.Save run_config.yaml from MegatronLM checkpoints
enhancementNew feature or requestNew feature or requestStatus: Open.- Status: Open.#5974 In NVIDIA/Megatron-LM;
[ENHANCEMENT] Fused forward+backward CUDA kernel for the CSA/HCA
Compressorgated pooling (THD path)enhancementNew feature or requestNew feature or requestwaiting-on-customerWaiting on the original author to respondWaiting on the original author to respondStatus: Open.#5968 In NVIDIA/Megatron-LM;Training crash with TE CUDA graphs
bugSomething isn't workingSomething isn't workingwaiting-on-customerWaiting on the original author to respondWaiting on the original author to respondStatus: Open.#5966 In NVIDIA/Megatron-LM;